mygo-cli

A MyGo meme image search tool with hybrid, semantic, and fuzzy search. Browse results interactively with fzf and copy images to clipboard.

It also includes a native macOS menu bar app. The app keeps the semantic model loaded in the background, so follow-up searches do not pay the model startup cost.

mygo.mp4

Download the macOS app

Download the latest Apple Silicon DMG from GitHub Releases, open it, and drag MyGo.app into Applications. The prebuilt app requires macOS 13 or later.

Requirements

• Apple Silicon Mac with macOS 13 or later for the prebuilt app
• Python 3.11 and uv for CLI/source development
• fzf and chafa for the optional interactive terminal interface

Note: macOS built-in Terminal.app and tmux render fzf preview poorly. For best experience, use iTerm2, Kitty, or Ghostty without tmux.

External Dependencies

brew install fzf chafa

Tool	Purpose
fzf	Interactive fuzzy finder
chafa	Terminal image preview

Installation

git clone https://github.com/shilohlee98/mygo-cli.git
cd mygo-cli
uv tool install .

If the tool is already installed and you want the new ONNX dependencies, reinstall it:

uv tool install --reinstall .

On first run, image data is fetched from the API and cached locally. MyGo downloads the prebuilt int8 multilingual-e5-small ONNX model and reuses it on later searches. Inference runs directly through ONNX Runtime and does not load PyTorch.

Usage

After installation, the mygo command is available globally:

# Interactive fzf mode (default)
mygo

# Search with a query directly
mygo haha

# Plain text output (no fzf)
mygo --no-fzf

# Plain text with query
mygo --no-fzf haha

Alternatively, run without installing globally:

uv run mygo

Performance toggles:

# Skip int8 quantization (uses substantially more memory)
MYGO_ONNX_QUANTIZATION=off mygo

macOS menu bar app

Run the app from source while developing:

uv run mygo-app

MyGo opens a native search window, stays visible in the Dock, and keeps running after the window is closed. Normal searches blend multilingual E5 semantic relevance with exact and fuzzy text matching. Prefix the query with /s for semantic-only results or /f for fuzzy-only results. Click Copy Image on any result to put it on the macOS clipboard.

Short English greetings are normalized only for semantic retrieval (hi and hello → 妳好). Hybrid text scoring still uses the original query, with English word boundaries so hi does not match inside names such as CRYCHIC.

Build a standalone app bundle:

./scripts/build_app.sh
open dist/MyGo.app

Build an installable DMG:

./scripts/build_dmg.sh

The outputs are dist/MyGo.app, dist/MyGo-<version>-<arch>.dmg, and a SHA-256 checksum. Build artifacts are intentionally ignored by Git. Local builds use an ad-hoc signature and are suitable for development on your own Mac.

For personal use, move the app to /Applications. To start it automatically after login, use MyGO → Open Login Items Settings… and add MyGo.app in macOS System Settings.

The image catalog, ONNX model, and embeddings are stored under:

~/Library/Application Support/MyGo/

The first launch needs network access and takes longer while these files are downloaded and generated. Later launches reuse them. If Application Support is not writable in the current launch context, MyGo reports that it is using a safe temporary cache instead of failing during ONNX setup.

Interactive Commands (--no-fzf mode)

• Type text — hybrid semantic and text search
• /s <query> — semantic-only search
• /f <query> — fuzzy search
• Enter a number — copy that image to clipboard
• q — quit

Acknowledgements

Image data provided by Miyago9267's MyGO-Searcher API.